Journal Archive

RDF is a metadata model proposed by World Wide Web Consortium (W3C). It has been used for conceptual description and modeling of information on web resources. It supports a variety of notations and data serializations. With RDF, semantic information on the Web can be processed by machines as well as by humans. While there are many other standards, RDF may be the simplest and most efficient to handle data and relationships between data.

To define vocabularies for RDF data, RDF Schema (RDFS) has also been developed by W3C. It is composed of various classes with certain properties for the structure of RDF data[3][4].

RDF data is expressed in three terms of subject, predicate, and object about resources, called triples. The subject and object denote resources, and the predicate expresses a relation between the subject and the object. They can be stored in and retrieved from a triplestore with the query language SPARQL[5].

SPARQL is a language to be used to query RDF triples and merge results from multiple data sources. It also enables Linked Open Data for the Semantic Web and enriches information by linking it to other semantic resources. Thus, data can be merged, shared, and reused in a more meaningful way[6].

SPARQL is analogous to SQL used to create, store, and retrieve structured data. SQL is suitable to access tables in a relational database, but SPARQL can access RDF triples. Although SPARQL was developed to combine diverse sources of data, it can be used to access relational data as well. Moreover, the SPARQL query may be constructed across a range of datasets so long as they are presented as a directed labelled graph. The results of SPARQL queries can be in the form of RDF triples[7].

Fig. 1 describes an example of a SPARQL query. The Block ① is PREFIX definition to specify URI of the related sites in the query. The Block ② is SELECT clause to retrieve variables. The Block ③ is WHERE clause to match with the triple patterns.

Fig. 1. 
Structure of SPARQL query

RDF triplestore is a special-purpose storage and retrieval system for the management of RDF triples. Most of the triplestores support the standard SPARQL as a query language. Users can define their own query patterns by combining provided primitives. SPARQL queries can be parsed and transformed with the dataset.

This paper adopts Apache Jena for the triplestore and SPARQL. Apache Jena provides API to store and extract data from the RDF triplestore. It also supports access control at the level of server and endpoint within a dataset[8][9].

This paper analyzed patent information structure to build an efficient storage system and then to alleviate the burden of finding related patents for a certain technical problem.

Table 1 shows the patent information structure in general. It consists of ten main categories, which are certificate, registration, application, references, applicant, assignee, inventor, examiner, general information, and classification. In each category, detailed attributes have been assigned accordingly.

Table 2 shows two of the ten categories that are certificate and registration category. The certificate category holds the certificates of correction, re-examination, PTAB trial, and supplemental exam. Registration category includes the international details of the registration number, registration date, publication date, and filing date.

Table 3 shows the next three categories, that are application, references, and applicant category. They include application type, application date, other references, reference by, foreign reference, applicant type, country, state, city, and name.

Table 4 shows two other categories that are the inventor and assignee category. They have the same attributes, which are name, country, state, and city, respectively. The inventor can be either one or more persons who have developed the invention. Whereas, the assignee is one with the property right to the patent.

Table 5 shows the last three categories that are classification, general, and examiner category. Classification category lists current CPC, current CPC class, and international classification. General category includes patent number, issue date, reissue data, title, claim(s), abstract, description, prior filing date, attorney/agent, and other attributes. The examiner category is divided into primary examiner and assistant examiner.

This paper also analyzed patent information structure from the Korea Intellectual Property Rights Information Service (KIPRIS). KIPRIS is a search site for intellectual property data in Korea.

Table 6 shows specific components of patent information within KIPRIS. It consists of an agent, applicant, inventor, and other attributes.

An architecture of RDF storage and query system is shown in Fig. 5. User can first construct its query and then accesses SPARQL server to process it. The system then processes the query conditions to match the datasets stored in the RDF storage. Next, the matched datasets are retrieved and analyzed thoroughly. Lastly, the result is returned to the user.

Fig. 5. 
Architecture of RDF storage and query system

The illustration in Fig. 6 describes how uncategorized raw data are analyzed, converted into RDF format, and finally uploaded in the RDF storage.

Fig. 6. 
Process of RDF storage

Fig. 7 shows the process of the SPARQL query to retrieve the matched result. First of all, the SPARQL server has to be run in the background. The construction of queries needs to be done in order to retrieve data from RDF storage.

Fig. 7. 
Process of RDF query

Fig. 8 and Fig. 9 show an example of a SPARQL query and the retrieved result, respectively. Retrieving information from RDF storage is done with the SPARQL query within the SPARQL server.

Fig. 8. 
SPARQL query for patent information

Fig. 9. 
Query result for applicant information

The followings are user interfaces that demonstrate the capabilities of the prototype system. Fig. 10 shows the datasets page that lists patent files in Turtle’s format (.ttl) on the RDF triplestore.

Fig. 10. 
Datasets page

Fig. 11 shows the query page where the SPARQL query can be constructed, and later the input query results will be displayed.

Fig. 11. 
Query page